Inference

First export the inference graph from the object detection step, and use Object-Detection/object_detection.py to generate the detection pickle file.

Along with the detection pickle, gather the classifier weights from the classification step, then proceed to generate either XMLs or bounding box overlays:

Predict and Generate XMLs

The predicted bounding boxes as well as the classifier grades are stored in the form of XML file. To see the annotated labels as predicted, use LabelImg.

Inference/predict_and_generate_xmls.py outputs the XML that corresponds to a matching image.

Predict and Draw Bounding Boxes with Probability

Inference/predict_and_draw_bounding_boxes_with_prob.py draws the bounding boxes as overlay on the original image. For machine predictions, it also fetches the probability of the inference confidence level along with the grade.